In Wikipedia's definition of Fréchet space, it is stated that a Fréchet space is a topological vector space that satisfies the following:

  • It is locally convex
  • Its topology can be induced by a translation-invariant metric
  • Any (hence every) translation-invariant metric inducing the topology is complete

1) My first question lies with the "hence every" clause. It is known that completeness is a property of the metric, not topology, the standard example being $(0,1)$ under the absolute value and arctan metrics. As such how is it that we can conclude completeness here for every translation-invariant metric inducing the topology?

2) Wikipedia gives yet another equivalent condition for the completeness property - namely, let $\{p_k\}_{k\in\mathbb{N}}$ denote a countable family of seminorms defining the topology of $X$; then the space is complete with respect to the family of seminorms (i.e. if $(x_n)$ is a sequence in $X$ which is Cauchy with respect to each seminorm $p_k$, then there exists $x\in X$ such that $(x_n)$ converges to $x$ with respect to each seminorm $p_k$). How does the completeness with respect to any translation-invariant metric imply the completeness with respect to any countable family of seminorms and vice versa?

Remark: I suppose an answer to Q2 will answer Q1.


Let $F$ be your Fréchet space.

Suppose that $d$ is a translation invariant metric that induced the topology of $F$ and that $F$ is complete under $d$. Let $\mathcal P=\{p_k\}_{k\in\mathbb N}$ be any family of seminorms inducing the topology of $F$. Let $$d'(x,y)=\sum_{k=1}^\infty 2^{-k}\,\frac {p_k(x-y)}{1+p_k (x-y)}$$ be the metric induced by $\mathcal P$. It is easy to check that this metric induces the same topology that $\mathcal P$, that is the topology of $F$.

Looking at the respective unit balls, we can find (since both metrics induce the same topology) $\alpha,\beta>0$ such that $$\alpha\,d(x,0)\leq d'(x,0)\leq \beta\,d(x,0).$$ Because the metrics are translation invariant, for any $x,y\in F$ we have $d(x,y)=d(x-y,0)$ and the same for $d'$ so $$\tag1\alpha\,d(x,y)\leq d'(x,y)\leq \beta\,d(x,y)$$for all $x,y\in F$. Thus a sequence will be complete for $d$ if and only if it is complete for $d'$.

As $(1)$ is an equivalence relation, we have that completeness is the same for any translation invariant metric that induces the topology for $F$, and also agrees with completeness with respect to any sequence of seminorms that induces the topology.

| cite | improve this answer | |
  • $\begingroup$ So in essence the argument is: every pair of equivalent translation-invariant metrics on a linear space is strongly equivalent (in the sense (1)) and strong equivalence of metrics preserves completeness (as opposed to simple topological equivalence). $\endgroup$ – Henno Brandsma Mar 3 '19 at 7:03
  • $\begingroup$ Yes, if by "equivalent" you mean that the two metrics determine the same open sets around $0$ (or, what is the same, they generate the same topology). $\endgroup$ – Martin Argerami Mar 3 '19 at 12:14
  • $\begingroup$ yes, that’s what I call topological equivalence. $\endgroup$ – Henno Brandsma Mar 3 '19 at 12:15
  • $\begingroup$ $d'$ isn't a metric (the series may diverge to $\infty$). The usual modification is $d'(x,y)= \sum\limits_{k=1}^\infty 2^{-k} \frac{p_k(x-y)}{1+p_k(x-y)}$. $\endgroup$ – Jochen Mar 4 '19 at 12:36
  • $\begingroup$ You are totally right. Editing right now. $\endgroup$ – Martin Argerami Mar 4 '19 at 13:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.